Optimization method and device for sparse antenna array by combining neural network and genetic algorithm
By integrating neural networks and genetic algorithms, the element position distribution of sparse antenna arrays is optimized, solving the problem of computational complexity in existing technologies, achieving efficient optimization of sparse antenna arrays, and improving algorithm performance and spatial resolution.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- PURPLE MOUNTAIN LAB
- Filing Date
- 2023-01-30
- Publication Date
- 2026-07-24
Smart Images
Figure CN116245173B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of antenna array technology, and in particular to a method and apparatus for optimizing sparse antenna arrays by integrating neural networks and genetic algorithms. Background Technology
[0002] In radar antenna arrays, a narrow beamwidth and low sidelobe level are often required to improve spatial resolution and other performance indicators. To achieve narrow beamwidth and low sidelobe level, antenna elements can be sparsely arranged across the array aperture to create a sparsely distributed antenna array. This reduces costs while increasing the spacing between elements, thus enlarging the antenna aperture and resulting in a narrower beamwidth.
[0003] There are many types of optimization algorithms for sparse antenna arrays, such as genetic algorithms, particle swarm optimization and its improved algorithms, hybrid algorithms combining simulated annealing, differential evolution algorithms, and so on. Among them, using genetic algorithms to optimize sparse antenna arrays requires calculating the antenna diagram and then using the antenna diagram to calculate the fitness function to determine the optimal solution, which is a relatively tedious and complex process. Summary of the Invention
[0004] To address the problems existing in the prior art, this invention provides a method and apparatus for optimizing sparse antenna arrays that integrates neural networks and genetic algorithms.
[0005] In a first aspect, the present invention provides a sparse antenna array optimization method integrating neural networks and genetic algorithms, comprising:
[0006] Initialize the genetic algorithm model; in the genetic algorithm model, the element position distribution of the sparse antenna array is used as the gene of an individual in the population, the absolute value of the weighted sum of the sidelobe peak levels of the sparse antenna array under multiple set beam directions is used as the fitness value, and the sparse distribution rate is used as the constraint condition.
[0007] The fitness values of different individuals in the initial population are determined based on a neural network model. The neural network model includes a first neural network model, which is trained based on the genes of the sample individuals and the antenna parameter data corresponding to the genes of the sample individuals. The antenna parameter data includes at least the data of the antenna pattern.
[0008] Determine whether the set termination condition has been met. If the set termination condition has been met, output the optimized result of the element position distribution of the sparse antenna array. Otherwise, if the set termination condition has not been met, update the population based on the genetic operator and the catastrophe operator, and determine the fitness value of different individuals in the updated population based on the neural network model. Repeat the steps of updating the population and determining the fitness value of different individuals in the updated population until the set termination condition is met, and output the optimized result of the element position distribution of the sparse antenna array.
[0009] Optionally, the antenna parameter data may also include data on antenna S-parameters and antenna gain.
[0010] Optionally, the neural network model further includes a second neural network model, which stores the fitness values associated with the genes of historical individuals.
[0011] Optionally, the second neural network model is a time-delayed neural network model, wherein each hidden layer of the time-delayed neural network model contains a time-delay unit, which is used to store the fitness value associated with the genes of historical individuals.
[0012] Optionally, the genetic operator includes one or more of a selection operator, a crossover operator, and a mutation operator;
[0013] The selection operator is used to select individuals to be inherited by the next generation;
[0014] The crossover operator is used to generate new individuals through gene crossover between individuals;
[0015] The mutation operator is used to generate new individuals through gene mutations in existing individuals.
[0016] Optionally, the catastrophe operator is used to replace the N individuals with the smallest fitness values in the population with N randomly generated individuals every set number of iterations;
[0017] The value of N is determined based on the current iteration number and the preset maximum iteration number.
[0018] Optionally, the value of N is based on The result of the operation is determined after rounding;
[0019] Where N0 represents the preset number of catastrophic individuals, a represents the preset catastrophic scale parameter, gen represents the current iteration number, and MAXGEN represents the preset maximum iteration number.
[0020] Secondly, the present invention also provides a sparse antenna array optimization device that integrates neural networks and genetic algorithms, comprising:
[0021] An initialization module is used to initialize the genetic algorithm model. In the genetic algorithm model, the element position distribution of the sparse antenna array is used as the gene of an individual in the population, the absolute value of the weighted sum of the sidelobe peak levels of the sparse antenna array under multiple set beam directions is used as the fitness value, and the sparse distribution rate is used as a constraint condition.
[0022] A determination module is used to determine the fitness values of different individuals in the initial population based on a neural network model; the neural network model includes a first neural network model, which is trained based on the genes of the sample individuals and the antenna parameter data corresponding to the genes of the sample individuals, and the antenna parameter data includes at least the data of the antenna pattern.
[0023] The output module is used to determine whether a set termination condition has been met. If the set termination condition has been met, the module outputs the optimized distribution result of the element positions of the sparse antenna array. Otherwise, if the set termination condition has not been met, the module updates the population based on the genetic operator and the catastrophe operator, and determines the fitness values of different individuals in the updated population based on the neural network model. The steps of updating the population and determining the fitness values of different individuals in the updated population are repeated until the set termination condition is met, and the optimized distribution result of the element positions of the sparse antenna array is output.
[0024] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the sparse antenna array optimization method integrating neural networks and genetic algorithms as described in the first aspect above.
[0025] Fourthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the sparse antenna array optimization method integrating neural networks and genetic algorithms as described in the first aspect above.
[0026] The present invention provides a method and apparatus for optimizing sparse antenna arrays by integrating neural networks and genetic algorithms. The fitness values of different individuals in each generation of the genetic algorithm model are determined by the neural network model, without the need for complex calculation formulas or simulation processes to calculate the fitness values. This improves the efficiency of genetic algorithm calculation and enhances the algorithm performance of sparse antenna arrays. Attached Figure Description
[0027] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0028] Figure 1 A flowchart illustrating the sparse antenna array optimization method that integrates neural networks and genetic algorithms provided by this invention;
[0029] Figure 2A schematic diagram of the structure of the first neural network model provided by the present invention;
[0030] Figure 3 This is a schematic diagram of the structure of the time-delay neural network model provided by the present invention;
[0031] Figure 4 A flowchart illustrating the optimization method for sparsely distributed antenna arrays provided by the present invention;
[0032] Figure 5 The training flowchart of the multi-parameter node neural network provided by the present invention;
[0033] Figure 6 The optimized normalized orientation pattern provided by this invention;
[0034] Figure 7 The optimized array antenna element distribution diagram provided by this invention;
[0035] Figure 8 A schematic diagram of the sparse antenna array optimization device that integrates neural networks and genetic algorithms provided by the present invention;
[0036] Figure 9 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation
[0037] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0038] In theory, an antenna consisting of a single antenna element can complete the task of transmitting and receiving electromagnetic waves. However, in practical applications, antennas are often required to have strong directivity and high gain. Sometimes, the antenna beam also needs to be able to scan and have a certain shape. In such cases, it is necessary to use multiple antenna elements arranged in a certain way to form an array antenna.
[0039] Furthermore, radar antenna arrays often require narrow beamwidths and low sidelobe levels to improve spatial resolution and other performance indicators. To achieve narrow beamwidths and low sidelobe levels, one approach is to increase the number of antenna array elements. This not only increases cost but also system complexity. Another approach is to sparsely arrange the antenna elements across the array aperture. This reduces cost while increasing the element spacing, thus enlarging the antenna aperture and resulting in a narrower beamwidth.
[0040] Compared to periodic arrays, in aperiodic arrays (sparse arrays), the antenna elements are sparsely arranged, the main lobe width of the antenna pattern remains basically unchanged, the mutual coupling effect between array elements is reduced, the sidelobe level is lowered, the feeding network is simplified, and the cost and failure rate of the antenna system are reduced.
[0041] There are many types of optimization algorithms for sparse antenna arrays, such as genetic algorithms, particle swarm optimization and its improved algorithms, hybrid algorithms combining simulated annealing, differential evolution algorithms, and so on.
[0042] This invention improves upon the traditional genetic algorithm by incorporating a neural network algorithm, thereby enhancing the performance of the sparse antenna array optimization algorithm.
[0043] Figure 1 The flowchart illustrates the sparse antenna array optimization method integrating neural networks and genetic algorithms provided by this invention. Figure 1 As shown, the method includes:
[0044] Step 100: Initialize the genetic algorithm model; In the genetic algorithm model, the element position distribution of the sparsely distributed antenna array is used as the gene of an individual in the population, the absolute value of the weighted sum of the sidelobe peak levels of the sparsely distributed antenna array under multiple set beam directions is used as the fitness value, and the sparse distribution rate is used as the constraint condition.
[0045] Step 101: Based on the neural network model, determine the fitness values of different individuals in the initial population; the neural network model includes a first neural network model, which is trained based on the sample individual genes and the antenna parameter data corresponding to the sample individual genes, and the antenna parameter data includes at least the antenna pattern data.
[0046] Step 102: Determine whether the set termination condition has been met. If the set termination condition has been met, output the optimized result of the element position distribution of the sparse antenna array. Otherwise, if the set termination condition has not been met, update the population based on the genetic operator and the catastrophe operator, and determine the fitness value of different individuals in the updated population based on the neural network model. Repeat the steps of updating the population and determining the fitness value of different individuals in the updated population until the set termination condition is met, and output the optimized result of the element position distribution of the sparse antenna array.
[0047] Specifically, the first step is to initialize the genetic algorithm model, which involves setting the initial parameters required for the model. These initial parameters include the genes of different individuals in the initial population, the fitness function, constraints, and termination conditions.
[0048] In order to obtain the optimal element position distribution of a sparse antenna array that meets certain conditions, in this invention, the element position distribution of different sparse antenna arrays is used as the genes of different individuals in the initial population; the absolute value of the weighted sum of the sidelobe peak levels of the sparse antenna array under multiple set beam directions is used as the fitness value (i.e. the value of the fitness function); and the set sparseness ratio is used as a constraint condition (for example, the sparseness ratio is set to 30%).
[0049] In one implementation, the fitness function F is formulated as follows:
[0050]
[0051]
[0052] In the formula, This is the array pattern of a sparsely distributed antenna array; m represents the elevation label of the array element; n represents the azimuth label of the array element; M represents the total number of different elevation labels; N represents the total number of different azimuth labels. This represents the radiation pattern of the array element with elevation symbol m and azimuth symbol n; λ represents the wavelength; d n d represents the azimuth distance of the array element with azimuth label n; m This indicates the elevation distance of the array element with the azimuth label m; For the initial beam pointing; g mn S1, S2, S3 represent the matrix of element positions with elevation symbol m and azimuth symbol n; MSLL represents the weighted sum of the sidelobe peak levels pointing downwards from multiple set beams; S1, S2, S3 n These represent different beam directions in the array pattern (i.e., corresponding to different beam directions respectively). Side lobe space at time; These represent different beam directions in the array pattern (i.e., corresponding to different beam directions respectively). The maximum sidelobe level at (A1, A2, A) is... n These are weighting coefficients set according to the actual optimization objective. The optimization objective is to find the element position distribution of the sparse antenna array corresponding to the array pattern with the largest fitness value.
[0053] The fitness value of each individual can be determined based on a neural network model, which may include a first neural network model.
[0054] When using the first neural network model to determine the fitness value of an individual, the individual's genes can be input into the first neural network model, the first neural network model outputs the array pattern of the antenna array corresponding to the individual, and then the fitness value of the individual can be determined based on the array pattern output by the first neural network model.
[0055] Optionally, the antenna parameter data used for training the first neural network model may also include data on antenna S-parameters and antenna gain. The antenna S-parameters can also be replaced with antenna standing wave ratio (SWR).
[0056] Figure 2 A schematic diagram of the structure of the first neural network model provided by the present invention is shown below. Figure 2 As shown, the first neural network model includes an input layer, a hidden layer, and an output layer. Antenna parameter data is input from the input layer. Figure 2 x1, x2, ..., x n After calculation in the hidden layer, the calculated antenna pattern is finally output from the output layer. Figure 2 In Antenna S-parameters Figure 2 S in the middle) and antenna gain ( Figure 2 (G in the middle).
[0057] The first neural network model is trained using data on antenna pattern, antenna S-parameters, and antenna gain. The trained first neural network model can simultaneously output the antenna pattern, antenna S-parameters, and antenna gain for each individual. Thus, it can also comprehensively evaluate individuals in the population based on conditions such as antenna S-parameters and antenna gain, and obtain a more suitable distribution of array element positions for the sparsely distributed antenna array.
[0058] Termination conditions can be that the fitness value of an individual reaches a pre-set optimization target value, or that the genetic algorithm model has been updated a preset number of times.
[0059] When a set termination condition is met, the optimized element position distribution of the sparse antenna array can be output. For example, when the fitness value of an individual reaches a preset optimization target value, the element position distribution of the sparse antenna array corresponding to that individual is output; or, when the number of model updates reaches a preset number of iterations, the element position distribution of the sparse antenna array corresponding to the individual with the highest fitness value in the current population is output.
[0060] If the set termination condition is not met, the population can be updated based on genetic operators and catastrophe operators, and the fitness values of different individuals in the updated population can be determined based on the neural network model. The steps of updating the population and determining the fitness values of different individuals in the updated population are repeated until the set termination condition is met, and the optimized distribution of the element positions of the sparse antenna array is output.
[0061] The sparse antenna array optimization method integrating neural networks and genetic algorithms provided by this invention determines the fitness value of different individuals in each generation of the genetic algorithm model through a neural network model. This eliminates the need for complex calculation formulas or simulation processes to obtain the fitness value, thereby improving the efficiency of genetic algorithm calculation and enhancing the algorithm performance of sparse antenna arrays.
[0062] Optionally, the neural network model may also include a second neural network model for storing fitness values associated with genes of historical individuals.
[0063] Specifically, the neural network model may also include a second neural network model, which may be placed before the first neural network model, and is used to store the fitness values associated with the genes of historical individuals.
[0064] The individual's gene is input into the second neural network model. If the second neural network model stores the fitness value associated with the individual's gene, it means that the fitness value associated with the individual's gene has been calculated, and the fitness value associated with the individual's gene can be directly output. If the second neural network model does not store the fitness value associated with the individual's gene, the second neural network model is skipped, and the individual's gene is directly input into the first neural network model, through which the individual's fitness value is determined.
[0065] The second neural network model determines whether there is duplicate calculation for the individual whose fitness value is to be calculated. If the fitness value of the individual has been calculated before, the stored fitness value associated with the individual's genes can be directly output, thereby reducing the computational load of the subsequent first neural network model and improving the computational efficiency and performance of the optimization algorithm for the sparsely distributed antenna array.
[0066] Optionally, the second neural network model is a time-delayed neural network model, in which each hidden layer contains a time-delay unit, which is used to store the fitness values associated with the genes of historical individuals.
[0067] Specifically, the second neural network model can be a Time Delay Neural Network (TDNN) model. Figure 3 This is a schematic diagram of the structure of the time-delay neural network model provided by the present invention, as shown below. Figure 3 As shown, each hidden layer of the time-delay neural network contains a time-delay unit, which is used to store the fitness values associated with the genes of historical individuals.
[0068] The time-delay neural network model adds a delay unit to each non-output layer of the feedforward network to record the most recent activity values of neurons. At time t, the activity value of neurons in layer s depends on the activity values of neurons in layer (s-1) over the most recent K time steps, i.e.:
[0069]
[0070] in M represents the activity value of neurons in layer s at time t. s This refers to the number of neurons. Through delay units, the feedforward network acquires the ability of short-term memory.
[0071] Using a time-delayed neural network model as a second neural network model allows the storage of fitness values associated with historical individual genes through delay units. If the fitness value of an individual has already been calculated, the stored fitness values associated with that individual's genes can be directly output, thereby reducing the computational load of the subsequent first neural network model and improving the computational efficiency and performance of the optimization algorithm for the sparsely distributed antenna array.
[0072] Optionally, the genetic operator includes one or more of the selection operator, crossover operator, and mutation operator;
[0073] Among them, the selection operator is used to select individuals to be inherited by the next generation;
[0074] The crossover operator is used to generate new individuals by crossing genes between individuals;
[0075] Mutation operators are used to generate new individuals through genetic mutations in existing individuals.
[0076] Specifically, the genetic operators used to update the population may include one or more of the selection operator, crossover operator, and mutation operator.
[0077] The selection operator is used to select individuals to be passed on to the next generation, that is, to select individuals with high fitness (i.e., large fitness values) to be passed on to the next generation; the crossover operator is used to generate new individuals through gene mutation, that is, to replace and recombine parts of the structure of two parent individuals to generate new individuals; the mutation operator is used to generate new individuals through gene mutation, that is, to modify certain gene values on the genes of an individual.
[0078] By introducing different types of genetic operators, new individuals can be generated through gene crossover and gene mutation, thereby increasing the population diversity and search capability of the genetic algorithm. By selecting individuals to inherit to the next generation, the genetic algorithm can be accelerated to converge to the optimal solution.
[0079] Optionally, the catastrophe operator is used to replace the N individuals with the smallest fitness values in the population with N randomly generated individuals every set number of iterations;
[0080] The value of N is determined based on the current iteration number and the preset maximum iteration number.
[0081] Specifically, the catastrophe operator is used to simulate the situation in nature where a biological population experiences drastic changes in its environment, such as earthquakes or volcanic eruptions, which can lead to the extinction of a large number of individuals in the population. However, there will always be individuals with strong adaptability who can survive the drastic changes in the environment, forming adaptive radiation and constituting a species replacement in biological history.
[0082] The specific implementation process of introducing the catastrophe operator can be to replace the N individuals with the worst fitness (i.e. the smallest fitness value) in the population with N randomly generated individuals every set number of iterations, thereby improving the diversity of the population during the evolution process.
[0083] The value of N is determined based on the current iteration number and the preset maximum iteration number.
[0084] Alternatively, the value of N can be determined according to... The result of the operation is determined after rounding;
[0085] Where N0 represents the preset number of catastrophic individuals, a represents the preset catastrophic scale parameter, gen represents the current iteration number, and MAXGEN represents the preset maximum iteration number.
[0086] By introducing a catastrophe factor, the N individuals with the lowest fitness values in the population are replaced with N randomly generated individuals every certain number of iterations. This can improve the population diversity of the genetic algorithm, help it escape local optima, and increase the algorithm's search capability.
[0087] The following examples of application of the method in specific scenarios further illustrate the method provided by this invention.
[0088] Figure 4 The flowchart illustrates the optimization method for sparsely distributed antenna arrays provided by this invention. Figure 4 As shown, the specific implementation steps of this method are as follows:
[0089] Step 1: Based on the specific parameter requirements to be optimized, set the initial parameters that the optimization algorithm needs to set, including the initial population that meets the conditions, the number of individuals, and the number of iterations of the algorithm.
[0090] Step 2: Before calculating fitness, a neural network with memory function (i.e., the second neural network model) is used to determine whether there is any duplicate calculation for the individual to be calculated. If so, its fitness value is directly output; otherwise, the fitness value is calculated.
[0091] Step 3: Input the individual genes in the population into a multi-parameter node neural network (i.e., the first neural network model) to quickly calculate the array pattern corresponding to each individual.
[0092] Figure 5 The training flowchart of the multi-parameter node neural network provided by this invention is as follows: Figure 5 As shown, the specific operation is as follows:
[0093] First, training data is extracted from electromagnetic field full-wave simulation software (such as HFSS, CST, etc.). The collected data is deduplicated, standardized, and split into three types of data packets, which are used for training antenna S-parameters, gain, and radiation pattern, respectively.
[0094] Furthermore, the training of these three branches is carried out simultaneously, enabling the model to learn the nonlinear mapping relationship between several antenna variables and the coefficients of the transfer function. After training is complete, the model is evaluated using the separated training data. The model's quality is determined by comparing electromagnetic full-wave simulation data and predicted data, and the model is optimized through feedback adjustment.
[0095] Furthermore, a suitable interface is finally established to enable the calling of the model in order to return the prediction results.
[0096] Step 4: Calculate the fitness function value of all individuals in the population and sort them according to their fitness values. The formula for calculating the fitness value F is as follows:
[0097]
[0098]
[0099] In the formula, This is the array pattern of a sparsely distributed antenna array; m represents the elevation label of the array element; n represents the azimuth label of the array element; M represents the total number of different elevation labels; N represents the total number of different azimuth labels. This represents the radiation pattern of the array element with elevation symbol m and azimuth symbol n; λ represents the wavelength; d n d represents the azimuth distance of the array element with azimuth label n; m This indicates the elevation distance of the array element with the azimuth label m; For the initial beam pointing; g mn S1, S2, S3 represent the matrix of element positions with elevation symbol m and azimuth symbol n; MSLL represents the weighted sum of the sidelobe peak levels pointing downwards from multiple set beams; S1, S2, S3 n These represent different beam directions in the array pattern (i.e., corresponding to different beam directions respectively). The space of the sidelobe at that time; These represent different beam directions in the array pattern (i.e., corresponding to different beam directions respectively). The maximum sidelobe level at (A1, A2, A) is... n These are weighting coefficients set according to the actual optimization objective. The optimization objective is to find the element position distribution of the sparse antenna array corresponding to the array pattern with the largest fitness value.
[0100] Step 5: Associate individual characteristic genes with their fitness values and store the calculation results in a neural network with memory function.
[0101] Step 6: Update the fitness value and determine whether it has reached the optimization target value and the set maximum number of iterations. If the optimization target value or the number of iterations has been reached, output the optimal solution for the current generation; otherwise, perform genetic factor calculation and catastrophe operator, and the operation process returns to step 2.
[0102] Furthermore, genetic factors include selection factors, crossover factors, and variation factors.
[0103] Furthermore, the catastrophic factor is implemented by replacing the N individuals with the worst fitness in the population with N randomly generated individuals every five generations, thereby increasing the diversity of the population during the evolutionary process.
[0104] The simulation parameters for the implementation example are set as follows: the diameter of the circular aperture is R = 80λ, the sparsity ratio is 30%, the population size is 500, and the maximum number of iterations is 300. The optimization objective is: within a ±50° scanning range, the sidelobe electrical average is below -15dB.
[0105] The proposed optimization method for sparse antenna arrays, which integrates neural networks and genetic algorithms, was used to optimize the array.
[0106] Figure 6 The optimized normalized radiation pattern provided by this invention, such as Figure 6 As shown, the sidelobe electrical average is below -15dB within the ±50° scanning range of the beam.
[0107] Figure 7 The optimized array antenna element distribution diagram provided by this invention is as follows: Figure 7 As shown, it can be seen that the sparse distribution rate of a sparsely distributed antenna array is about 30% compared to a full array.
[0108] The final example verification shows that the sparse antenna array optimization method that integrates neural networks and genetic algorithms provided by this invention can achieve a scanning radiation pattern with low sidelobe levels in a phased array antenna by optimizing the position of array elements.
[0109] The following describes the sparse antenna array optimization device that integrates neural networks and genetic algorithms provided by the present invention. The sparse antenna array optimization device that integrates neural networks and genetic algorithms described below can be referred to in correspondence with the sparse antenna array optimization method that integrates neural networks and genetic algorithms described above.
[0110] Figure 8 A schematic diagram of the sparse antenna array optimization device integrating neural networks and genetic algorithms provided by the present invention is shown below. Figure 8 As shown, the device includes:
[0111] Initialization module 800 is used to initialize the genetic algorithm model. In the genetic algorithm model, the element position distribution of the sparse antenna array is used as the gene of the individual in the population, the absolute value of the weighted sum of the sidelobe peak levels of the sparse antenna array under multiple set beam directions is used as the fitness value, and the sparse distribution rate is used as the constraint condition.
[0112] The determination module 810 is used to determine the fitness values of different individuals in the initial population based on a neural network model. The neural network model includes a first neural network model, which is trained based on the genes of the sample individuals and the antenna parameter data corresponding to the genes of the sample individuals. The antenna parameter data includes at least the data of the antenna pattern.
[0113] The output module 820 is used to determine whether the set termination condition has been met. If the set termination condition has been met, it outputs the optimized result of the element position distribution of the sparse antenna array. Otherwise, if the set termination condition has not been met, it updates the population based on the genetic operator and the catastrophe operator, and determines the fitness value of different individuals in the updated population based on the neural network model. The steps of updating the population and determining the fitness value of different individuals in the updated population are repeated until the set termination condition is met, and the optimized result of the element position distribution of the sparse antenna array is output.
[0114] Optionally, the antenna parameter data may also include data on the antenna S-parameters and antenna gain.
[0115] Optionally, the neural network model may also include a second neural network model for storing fitness values associated with genes of historical individuals.
[0116] Optionally, the second neural network model is a time-delayed neural network model, in which each hidden layer contains a time-delay unit, which is used to store the fitness values associated with the genes of historical individuals.
[0117] Optionally, the genetic operator includes one or more of the selection operator, crossover operator, and mutation operator;
[0118] Among them, the selection operator is used to select individuals to be inherited by the next generation;
[0119] The crossover operator is used to generate new individuals by crossing genes between individuals;
[0120] Mutation operators are used to generate new individuals through genetic mutations in existing individuals.
[0121] Optionally, the catastrophe operator is used to replace the N individuals with the smallest fitness values in the population with N randomly generated individuals every set number of iterations;
[0122] The value of N is determined based on the current iteration number and the preset maximum iteration number.
[0123] Optionally, the value of N is based on The result of the operation is determined after rounding;
[0124] Where N0 represents the preset number of catastrophic individuals, a represents the preset catastrophic scale parameter, gen represents the current iteration number, and MAXGEN represents the preset maximum iteration number.
[0125] It should be noted that the device provided by the present invention can implement all the method steps implemented in the above method embodiments and can achieve the same technical effect. Therefore, the parts and beneficial effects that are the same as those in the method embodiments will not be described in detail here.
[0126] Figure 9 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 9 As shown, the electronic device may include a processor 910, a communications interface 920, a memory 930, and a communication bus 940, wherein the processor 910, the communications interface 920, and the memory 930 communicate with each other via the communication bus 940. The processor 910 can call logical instructions in the memory 930 to execute any of the sparse antenna array optimization methods integrating neural networks and genetic algorithms provided in the above embodiments.
[0127] Furthermore, the logical instructions in the aforementioned memory 930 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0128] It should be noted that the electronic device provided by the present invention can implement all the method steps implemented in the above method embodiments and can achieve the same technical effect. Therefore, the parts and beneficial effects that are the same as those in the method embodiments will not be described in detail here.
[0129] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform any of the above-described methods for optimizing sparse antenna arrays by fusing neural networks and genetic algorithms.
[0130] It should be noted that the non-transitory computer-readable storage medium provided by the present invention can implement all the method steps implemented in the above method embodiments and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiments and the beneficial effects will not be described in detail.
[0131] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0132] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for optimizing sparsely distributed antenna arrays by integrating neural networks and genetic algorithms, characterized in that, include: Initialize the genetic algorithm model; in the genetic algorithm model, the element position distribution of the sparse antenna array is used as the gene of an individual in the population, the absolute value of the weighted sum of the sidelobe peak levels of the sparse antenna array under multiple set beam directions is used as the fitness value, and the sparse distribution rate is used as the constraint condition. The fitness values of different individuals in the initial population are determined based on a neural network model. The neural network model includes a first neural network model, which is trained based on the genes of the sample individuals and the antenna parameter data corresponding to the genes of the sample individuals. The antenna parameter data includes at least the data of the antenna pattern. The algorithm determines whether a set termination condition has been met. If the condition is met, it outputs the optimized element position distribution result of the sparse antenna array. Otherwise, if the termination condition is not met, it updates the population based on genetic and catastrophe operators, and determines the fitness values of different individuals in the updated population based on the neural network model. This process is repeated until the termination condition is met, at which point the optimized element position distribution result of the sparse antenna array is output. The catastrophe operator is used to replace the N individuals with the smallest fitness values in the population with N randomly generated individuals every set number of iterations.
2. The sparse antenna array optimization method integrating neural networks and genetic algorithms according to claim 1, characterized in that, The antenna parameter data also includes data on antenna S-parameters and antenna gain.
3. The sparse antenna array optimization method integrating neural networks and genetic algorithms according to claim 1 or 2, characterized in that, The neural network model also includes a second neural network model, which is used to store the fitness values associated with the genes of historical individuals.
4. The sparse antenna array optimization method integrating neural networks and genetic algorithms according to claim 3, characterized in that, The second neural network model is a time-delayed neural network model. Each hidden layer of the time-delayed neural network model contains a time-delay unit, which is used to store the fitness values associated with the genes of historical individuals.
5. The sparse antenna array optimization method integrating neural networks and genetic algorithms according to claim 1, characterized in that, The genetic operators include one or more of the selection operator, crossover operator, and mutation operator; The selection operator is used to select individuals to be inherited by the next generation; The crossover operator is used to generate new individuals through gene crossover between individuals; The mutation operator is used to generate new individuals through gene mutations in existing individuals.
6. The sparse antenna array optimization method integrating neural networks and genetic algorithms according to claim 1, characterized in that, The value of N is determined based on the current iteration number and the preset maximum iteration number.
7. The sparse antenna array optimization method integrating neural networks and genetic algorithms according to claim 6, characterized in that, The value of N is based on The result of the operation is determined after rounding; in, This indicates the preset number of individuals affected by the disaster. a This indicates the preset disaster scale parameters. Indicates the current iteration number. This indicates the preset maximum number of iterations.
8. A sparse antenna array optimization device integrating neural networks and genetic algorithms, characterized in that, include: An initialization module is used to initialize the genetic algorithm model. In the genetic algorithm model, the element position distribution of the sparse antenna array is used as the gene of an individual in the population, the absolute value of the weighted sum of the sidelobe peak levels of the sparse antenna array under multiple set beam directions is used as the fitness value, and the sparse distribution rate is used as a constraint condition. A determination module is used to determine the fitness values of different individuals in the initial population based on a neural network model; the neural network model includes a first neural network model, which is trained based on the genes of the sample individuals and the antenna parameter data corresponding to the genes of the sample individuals, and the antenna parameter data includes at least the data of the antenna pattern. The output module is used to determine whether a set termination condition has been met. If the set termination condition has been met, it outputs the optimized element position distribution result of the sparse antenna array. Otherwise, if the set termination condition has not been met, it updates the population based on the genetic operator and the catastrophe operator, and determines the fitness values of different individuals in the updated population based on the neural network model. The steps of updating the population and determining the fitness values of different individuals in the updated population are repeated until the set termination condition is met, and the optimized element position distribution result of the sparse antenna array is output. The catastrophe operator is used to replace the N individuals with the smallest fitness values in the population with N randomly generated individuals every set number of iterations.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the sparse antenna array optimization method that integrates neural networks and genetic algorithms as described in any one of claims 1 to 7.
10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the sparse antenna array optimization method that integrates neural networks and genetic algorithms as described in any one of claims 1 to 7.